Full text loading...
-
Optimal Survey Design for Big Data
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, 76th EAGE Conference and Exhibition 2014, Jun 2014, Volume 2014, p.1 - 5
Abstract
Industrial geoscience is entering the age of ‘big data’, in which the data volumes routinely acquired for analysis are so large that they can no longer be processed by traditional workflows. How can we store and mine this deluge of information?
In this article, we focus on model-oriented design and analysis (MODA) -- the theory and practice of designing experiments to maximize the information expected in data observations. MODA is appealing because it is an optimization method, which offers confidence in the expected results, and because it can be applied before or after data acquisition, to either forecast the most informative data to acquire or to optimally select data from existing datasets. MODA reduces costs by increasing efficiency in either scenario, whether through data acquisition or data processing.
However, MODA is itself challenged by the computational demands of big data, and researchers are seeking ways to reduce its computational cost. One novel possibility is dimension reduction – especially if it is parallelizable. We develop a dimension reduction workflow for guided Bayesian survey design – a linearized MODA technique – that greatly reduces the computing cost of optimal survey design, and we demonstrate its utility on a real, industrial-scale marine seismic design problem.