27 February 2018

Design of Measurement Networks for Predictive Uncertainty Reduction

Science
  • Design a cost-effective measurement network to improve model prediction and reduce prediction uncertainty.
  • Develop a computationally efficient method to enhance the evaluation of model prediction uncertainties.
Impact
  • The MLMC-based measurement network design will guide data collection and help model development. 
  • Quantification of predictive uncertainties will provide actionable information to policymakers.
  • Next, we will apply the approach to a coupled land-atmosphere model to optimize the placement of new simulation points, so as to improve climate model predictions. 
Summary

Improving the understanding of subsurface systems and thus reducing prediction uncertainty requires the collection of data. As the collection of subsurface data is costly, it is important that the data collection scheme is cost-effective. Design of a cost-effective data collection scheme, i.e., data-worth analysis, requires quantifying model parameter, prediction, and both current and potential data uncertainties. Assessment of these uncertainties in large-scale stochastic subsurface hydrological model simulations using standard Monte Carlo (MC) sampling or surrogate modeling is extremely computationally intensive, sometimes even infeasible. In this work, we propose an efficient Bayesian data-worth analysis using a multilevel Monte Carlo (MLMC) method. Compared to the standard MC that requires a significantly large number of high-fidelity model executions to achieve a prescribed accuracy in estimating expectations, the MLMC can substantially reduce computational costs using multifidelity approximations. Since the Bayesian data-worth analysis involves a great deal of expectation estimation, the cost saving of the MLMC in the assessment can be outstanding. While the proposed MLMC-based data-worth analysis is broadly applicable, we use it for a highly heterogeneous two-phase subsurface flow simulation to select an optimal candidate data set that gives the largest uncertainty reduction in predicting mass flow rates at four production wells. The choices made by the MLMC estimation are validated by the actual measurements of the potential data, and consistent with the standard MC estimation. But compared to the standard MC, the MLMC greatly reduces the computational costs.

Contact
Dan Lu
Oak Ridge National Laboratory (ORNL)
Funding
Programs
  • Scientific Discovery through Advanced Computing
  • Terrestrial Ecosystem Science
Publications
Lu, D, D Ricciuto, and K Evans.  2018.  "An Efficient Bayesian Data-Worth Analysis Using a Multilevel Monte Carlo Method."  Advances in Water Resources 113: 223-235, doi:10.1016/j.advwatres.2018.01.024.